View
2
Download
0
Category
Preview:
Citation preview
Data-Centric Secure Computing
Dr. Emily Shen
MIT Lincoln Laboratory
5 March 2018
The Future of Advanced (Secure) Computing
This material is based upon work supported by the Assistant Secretary of Defense for Research and Engineering under Air Force Contract No. FA8721-05-C-0002 and/or FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Assistant Secretary of Defense for Research and Engineering.
DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited.
© 2018 Massachusetts Institute of Technology.
Data-Centric Secure Computing - 3EHS 03/05/18
Need for Secure Computing on Data
Internet of Things Medical ResearchCyber Threat SharingCloud Computing
Data-Centric Secure Computing - 4EHS 03/05/18
Data-Centric Secure Computing
FUTURE: DATA-CENTRIC
Storage (Data at Rest)
Computation(Data in Use)
Networking(Data in Transit)
Fine-grained access control self-enforced by
data
Only authorized computation allowed by
data
Based on data identifiers(content-centric
networking)
CURRENT: SYSTEM-CENTRIC
Storage(Data at Rest)
Computation (Data in Use)
Networking(Data in Transit)
Access control enforced by system; single point of
failure
All computation allowed by system
Based on systemidentifiers
(IP addresses)
Vision: Self-protecting data throughout data lifecycle in distributed systems
Data-Centric Secure Computing - 5EHS 03/05/18
Data-Centric Secure Computing Architecture
• Publish data• Request data
Secure ComputationEngines
Secure Data Capsule
Data
Name
Policies
Provenance
History of interaction• Data capsules• Functions
Content-CentricNetwork
Authorized uses• Read/write• Compute
Security requirements• Confidentiality• Integrity
HomomorphicEncryption
Multi-partyComputation
Enclave Processor
FunctionalEncryption
Data-Centric Secure Computing - 6EHS 03/05/18
Data-Centric Secure Computing for Medical Research
Hospital
Node
Controls
Cancer Research Task
Requests computation
Retrieves result
Researcher
Controls
Encrypts
Content-Centric Network
NodeCompute
NodeSecure Comp
Engines
Genomic Data
Matching Function
Matching Result
DNA Sequencer
Data-Centric Secure Computing - 7EHS 03/05/18
Secure Computation Example: Multi-Party Computation (MPC)
• MPC uses cryptography to emulate functionality and security of a trusted party– Confidentiality of inputs and outputs– Correctness of computation– Resilience to communication/party failures
Ideal World Real World
MPC
Data-Centric Secure Computing - 8EHS 03/05/18
1. Secret share inputs– Each party encodes private data, sends a share
to each party
– Shares completely hiding unless more than t shares are combined
2. Compute on secret shares– Addition uses only local computation
– Multiplication requires communication
3. Open output: Combine final shares to learn result
MPC Protocols
1 2 3
s2f2(x)
s1
f1(x)
s1 + s2
(f1 + f2)(x)
Party
Share
Secret sharing for threshold t = 1
MPC can compute any arbitrary function securely, can be optimized for specific applications
BGW – M. Ben-or, S. Goldwasser, A. Wigderson. Completeness Theorems for Non-Cryptographic Fault-Tolerant Distributed Computation. STOC 1988
Data-Centric Secure Computing - 9EHS 03/05/18
Example: Optimizing MPC Sorting Protocols
0
120
240
360
480
600
720
840
960
1080
1200
12 24 48 96 192 384 768 1536 3072
Ru
nn
ing
Tim
e (
seco
nd
s)
Total Number of Input Values
Linear MergeO(n) ops,
O(n) rounds
Pairwise Comparison SortO(n2) ops,
O(1) rounds
Batcher SortO(n log2n) ops, O(log2n) rounds
Odd-Even MergeO(n log n) ops, O(log n) rounds
3 parties on Linux VMs w/Ubuntu 14.04, 2-vCPU at 2.0 GHz, 8 GB RAM/disk, 10
gigabit network, using Sepia framework*
Optimal MPC sorting protocol depends on preconditions and number of inputs
Data-Centric Secure Computing - 10EHS 03/05/18
Research Challenges
Data Provenance
• Truncation-resistant provenance store
• Provenance analytics
Security Policies
• Rich policy representation formats
• Combining policies on data from multiple owners
Secure Data Capsule
• Transformation of data to match protections specified by policy
• Integration with policy and provenance
• Automatic selection and composition of techniques
• Integration with policy and provenance
SecureComputation
• Secure resource discovery
• Resilience against malicious nodes
Content-CentricNetworking
Data-Centric Secure Computing - 11EHS 03/05/18
Summary
• Data-centric secure computing shifts paradigm from protecting large systems to protecting data
• Data protected at rest, in transit, and in use with respect to expressive policies
• Vision requires integrated architecture and component technologies: cryptographically secure storage and computation, policy, data provenance, content-centric networking
• Interested in your ideas for applications and collaboration
Recommended